A Study Of Correlation-based Initial Model Building Method For Full Waveform Inversion

In seismic data processing and interpretation,velocity is an indispensable important parameter,it plays an important role in the description of subsurface structure and reservoir prediction.In recent years,the degree of exploration and development of oil and gas fields has been increasing day by day,exploration targets have gradually transferred from simple oil-bearing structures into complex structures and subtle reservoirs,the requirement for velocity model rebuilding is getting higher in seismic exploration.Traditional velocity rebuilding methods,such as traveltime tomography and migration velocity analysis,cannot meet the requirements of seismic data processing and interpretation under complex geological conditions in accuracy and resolution.Full waveform inversion is a kind of velocity model rebuilding method that makes full use of all the information in prestack seismic wave field.It obtains the corresponding subsurface velocity model by using an optimization method to match the simulated records and observed records and minimize the error function between them.Full waveform inversion can reveal the details of subsurface structure even under complex geological conditions,and it has higher resolution and accuracy than traditional velocity rebuilding methods.In this paper,we studied the theory of two-dimensional full waveform inversion in time domain,deduced the high-order finite difference scheme of acoustic wave equation and the gradient formula of objective function,and analysed the stability of full waveform inversion.The conventional full waveform inversion using 2l norm objective function is highly nonlinear,with the complexity of the model parameter variation,the inversion is unstable and is easily trapped in local minima during the iteration process.In order to make the inversion better converged to the global minimum,it is usually necessary to provide a sufficiently accurate initial velocity model or seismic data that contain low frequency information with high signal-to-noise ratio.To solve this problem,this paper proposed a correlation-based initial model building method,which aims at providing a good initial model for conventional FWI and thus improving the stability of the inversion.This method uses a correlation-based objective function to measure the similarity between the simulated record and the observed record,it emphasizes the phase matching between the seismic records and has better linearity,so it is less likely to fall into local minima.The correlaton-based FWI can rebuild the macro structure of the model,which can be used as an initial model for conventional FWI and improve the accuracy and stability of the inversion.Considering the huge computational cost of FWI,this paper proposed an efficient full waveform inversion method in time domain,introducing the blended source technology to improve computational efficiency.The interaction between single sources that make up a supershot will produce serious crosstalk noise,so we proposed a dynamic random encoding strategy to depress the noise.This method randomly selects several sources to make up a supershot after every ten iterations,and randomly encodes the amplitude and phase of each source.Model tests show that the blended source FWI based on dynamic random encoding strategy significantly improve the efficiency of FWI in the promise of inversion quality.In order to verify the inversion results of the proposed method in the absence of low-frequency components in seismic data,the high-pass filter was used to filter out the low-frequency components of the seismic wavelet.Results show that the proposed method is less dependent on low-frequency information than the conventional method.In addition,the noise test was carried out to further verify that the proposed method has good performance even when the seismic data contain noise.